[cffi-devel] [PATCH] 64-bit support for Allegro, and allocate stack objects in the static area

[John Fremlin]

"Luís Oliveira" <luismbo at gmail.com> writes:

> On Fri, Jan 9, 2009 at 12:51 AM, John Fremlin <jf at msi.co.jp> wrote:
>>> The page <http://www.franz.com/support/documentation/8.1/doc/foreign-functions.htm>
>>> documents the :FOREIGN-ADDRESS type.
>>
>> I kept trying to point out that this was being used by cffi as a *Lisp*
>> type not a *foreign-type*. It is not a Lisp type.
>
> Oh, yes, that part was clearly broken. But it was being used a foreign
> type as well, but even then it doesn't seem to work in all contexts.

Yes, I don't think it is even a "proper" foreign type, it is a bit like :void.

>> The other change from my patch was about allocating the stack objects in
>> a space that will not be moved by the GC.
>
> I've applied that already, thanks! IIUC, :FOREIGN-STATIC-GC is only
> relevant when WITH-STACK-FOBJECT fails to stack allocate, in which
> case it actually allocates on the heap.

http://www.franz.com/support/documentation/8.1/doc/operators/ff/with-stack-fobject.htm

        The allocation argument defaults to :foreign, and can be one of
        :c, :aligned, :lisp, :lisp-short, :foreign, or
        :foreign-static-gc. Note that if allocation is :c or :aligned,
        it is not stack-allocated, but instead is allocated and
        deallocated at the appropriate places within the form.

This is actually a little misleading, as it goes on to say

     Otherwise, the object is allocated as specified. In this case, if
     the allocation requires explicit de-allocation, it is the
     responsibility of the application to de-allocate the object.

http://www.franz.com/support/documentation/8.1/doc/operators/ff/with-static-fobject.htm

with-static-fobject on the other hand behaves in a much more sensible
way. It will try to allocate on the stack but if it can't it will
unwind-protect a deallocation.

> Hmm, now that I read the documentation more carefully, it seems that
> if WITH-STACK-FOBJECT allocates on the heap then it won't deallocate
> the object and we should be using WITH-STATIC-FOBJECT instead. Does
> that sound right? If it does, I'll apply the attached patch during the
> weekend or so.

Why do you still need the

   (cond ((and (constantp size) (<= (eval size) ff:*max-stack-fobject-bytes*))

This was just to check whether the with-stack-fobject would actually
allocate the thing on the stack (if it did not then we would have to
free it by ourselves, and so we would not use with-stack-fobject at
all). 

You can just use with-static-fobject for everything now and let Allegro
decide whether it can allocate on the stack?